Identification and prediction of novel classes of long-term disease trajectories for patients with juvenile dermatomyositis using growth mixture models

OBJECTIVES: Uncertainty around clinical heterogeneity and outcomes for patients with JDM represents a major burden of disease and a challenge for clinical management. We sought to identify novel classes of patients having similar temporal patterns in disease activity and relate them to baseline clinical features.

METHODS: Data were obtained for n = 519 patients, including baseline demographic and clinical features, baseline and follow-up records of physician's global assessment of disease (PGA), and skin disease activity (modified DAS). Growth mixture models (GMMs) were fitted to identify classes of patients with similar trajectories of these variables. Baseline predictors of class membership were identified using Lasso regression.

RESULTS: GMM analysis of PGA identified two classes of patients. Patients in class 1 (89%) tended to improve, while patients in class 2 (11%) had more persistent disease. Lasso regression identified abnormal respiration, lipodystrophy and time since diagnosis as baseline predictors of class 2 membership, with estimated odds ratios, controlling for the other two variables, of 1.91 for presence of abnormal respiration, 1.92 for lipodystrophy and 1.32 for time since diagnosis. GMM analysis of modified DAS identified three classes of patients. Patients in classes 1 (16%) and 2 (12%) had higher levels of modified DAS at diagnosis that improved or remained high, respectively. Patients in class 3 (72%) began with lower DAS levels that improved more quickly. Higher proportions of patients in PGA class 2 were in DAS class 2 (19%, compared with 16 and 10%).

CONCLUSION: GMM analysis identified novel JDM phenotypes based on longitudinal PGA and modified DAS.